Research map / Adversarial robustness and security / Attacks on text and graphs
Attacking CNNs in Histopathology with SNAP: Sporadic and Naturalistic Adversarial Patches (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes SNAP, a sporadic, naturalistic multi-patch adversarial attack designed for histopathology images, which drops a fine-tuned ResNet50's metastasis-detection accuracy from 96.28 to 69.0.
Problem
CNNs used for histopathological cancer detection are vulnerable to adversarial patch attacks; existing patch attacks use a single large, visible patch and defenses target single patches.
Approach
A multi-piece adversarial patch whose pieces are spread sporadically over the image in cell-like shapes, with colors constrained to the image's pink palette and initialized from the median pixel color, trained jointly with a masked patch formulation.
Key results
- SNAP reduces ResNet50 accuracy on HCD metastasis detection from 96.28 to 69.0, a drop of about 27 points.
- The fine-tuned ResNet50 reaches 96.28 training accuracy and 89.35 validation accuracy before attack.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Apply trained SNAP targeting class 0 | ResNet50 · HCD | accuracy: 96.28 to 69.0 | Hurt |
Limitations
- Evaluated only on a single CNN (ResNet50) and a single dataset.
- No comparison with baseline patch attacks or defenses.
- Naturalistic appearance is claimed but not quantified by a human or perceptual study.
Experimental setup
Datasets: Histopathologic Cancer Detection (HCD) dataset
Benchmarks: Histopathologic Cancer Detection (HCD) dataset
Metrics: accuracy
Builds on: adversarial patch attacks
Closest papers by meaning
- Prompt2Perturb (P2P): Text-Guided Diffusion-Based Adversarial Attack on Breast Ultrasound Images (CVPR 2025 · similarity 0.84)
- NaFV-Net: An Adversarial Four-view Network for Mammogram Classification (AAAI 2025 · similarity 0.83)
- Beyond ViT Tokens: Masked-Diffusion Pretrained Convolutional Pathology Foundation Model for Cell-Level Dense Prediction (NeurIPS 2026 · similarity 0.83)
- A Textbook Remedy for Domain Shifts: Knowledge Priors for Medical Image Analysis (NeurIPS 2024 · similarity 0.83)
- Targeted Activation Penalties Help CNNs Ignore Spurious Signals (AAAI 2024 · similarity 0.82)
- CPath-Omni: A Unified Multimodal Foundation Model for Patch and Whole Slide Image Analysis in Computational Pathology (CVPR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.